Abstract 108P
Background
To improve the detection of early emphysema in a low dose CT (LDCT) lung cancer screening dataset.
Methods
We selected 352 participants from a regional community health center lung cancer screening dataset. Unenhanced low dose CTs were performed using Definition AS (Somatom Definition AS 64), at inspiration, in spiral mode at 120 kVp and 35 mAs. Images were reconstructed with B30f kernel at 2.0/1.0 mm thickness. AI based post-processing software, Aview (Coreline soft, v. 1.0.40), was used to automatically segment the lungs while excluding the pulmonary vessels and bronchus. Emphysema was quantified by voxel counting below a Hounsfield unit (HU) threshold. A wide range of thresholds from −900 to −1024 HU were used. Three readers including one experienced general radiologist (reader A) and two trainees in thoracic radiology (reader B, C) read the CT images by evaluating the emphysema according to the Fleischner criteria. Inter-reader agreement was evaluated using statistical analysis with Cohen's Kappa. Spearman analysis and ROC (Receiver operating characteristic) curve were used to assess the correlation between quantified emphysema under different HU thresholds and reader visual evaluation.
Results
184 (52.3%), 146 (41.5%), and 185 (52.6%) of the cases were classified respectively by reader A, reader B and reader C as positive for emphysema. All readers showed high agreement in diagnosis of the cases. The p value of Spearman analysis is less than 0.05, demonstrating a statistically significant correlation between emphysema volume and the visual classification under different thresholds. The optimal HU threshold was −1000 UH for all readers. The area under the curve (AUC) was 0.799 (95%CI: 0.751–0.847) for reader A, 0.797 (95%CI: 0.751–0.843) for reader B, and 0.785 for reader C (95%CI: 0.738–0.832).
Conclusions
Threshold of −1000 HU was determined to be optimal for the early detection of emphysema in LDCT lung cancer screening dataset. HU threshold optimization for automatic early emphysema detection by CT is indicated.
Legal entity responsible for the study
The authors.
Funding
Supported by the Royal Netherlands Academy of Arts and Sciences (grant PSA_SA_BD_01). Y.M. supported by the China Scholarship Council (CSC no. 202008440409).
Disclosure
All authors have declared no conflicts of interest.